RamanPFN: learning from Raman spectral structure with a tabular foundation model

Spectral Parsimony: A Novel Interpretational Approach to Feature Selection in Vibrational Spectroscopy

Pic2Spec: Generative Modeling Reconstructs Single Cell Raman Fingerprints from Brightfield Images

Single-cell molecular characterization remains a bottleneck in scalable biological analysis because of labeling requirements, limited multiplexing, and reagents that perturb physiology. Raman spectroscopy addresses these limits by providing chemically specific, label-free vibrational fingerprints, but long acquisition times and specialized instruments restrict high-throughput use. Here, we overcome this barrier by showing that spectral fingerprints can be reconstructed from brightfield microscopy using generative modeling. We introduce Pic2Spec, a framework that learns a shared latent biochemical representation linking image morphology to vibrational spectral structure, enabling virtual Raman spectroscopy without hardware. We validate Pic2Spec across mammalian and bacterial cells, generating high-fidelity spectra that reproduce measured Raman fingerprints with 98% cosine similarity and Pearson correlations of ~95%, while preserving biochemical peaks and population distributions. Beyond spectral similarity, Pic2Spec provides molecular-level resolution in bacterial systems: generated spectra discriminate mutation-driven transgenic states and predict GFP expression with accuracy approaching true Raman measurements, outperforming conventional image analysis by 20%. These findings establish Pic2Spec as a first demonstration of chemically informative virtual molecular fingerprinting from brightfield images, complementing slow, hardware-intensive spectroscopy with computational inference. By redefining microscopy as an inference-enabled molecular profiling platform, Pic2Spec democratizes label-free biochemical phenotyping and overcomes the hardware and time constraints that have confined spectroscopy to specialized laboratories. This enables high-throughput molecular analysis for clinical diagnostics, screening, and monitoring at the scale and accessibility of standard microscopy.

Sequential Learning of Raman Signatures From Attenuated Spectral Inputs Using Deep Neural Networks

Raman spectroscopy suffers from inherently weak scattering, especially in organic and biological samples, limiting its utility for low‐concentration analysis. Despite improvements in instrumentation, achieving an adequate signal‐to‐noise ratio in dilute aqueous systems remains challenging. To overcome this limitation, a deep learning (DL) framework was developed that performs spectral‐to‐spectral (S2S) regression rather than relying on classification or qualitative analysis. The framework reconstructs feature‐rich Raman spectra from weak, noise‐dominated inputs by learning concentration‐dependent spectral transformations across the full frequency range, thereby preserving peak integrity and enhancing signal quality. Five chemically diverse compounds spanning 1 nM to 2 M concentrations were used to train and evaluate multiple architectures, including convolutional neural networks (CNNs), recurrent neural networks (RNNs). RNNs variant, BiLSTM demonstrated the highest accuracy in capturing long‐range spectral dependencies, achieving superior R2, RMSE, and SSIM metrics. Robustness was validated using Monte Carlo uncertainty quantification and a leave‐one‐sample‐out (LOSO) strategy to assess generalization across unseen samples. This study presents a cost‐effective and computation‐driven strategy to simulate the rich Raman spectral features where conventional measurements fail. Beyond spectral reconstruction, the proposed approach can be extended to real‐time biomedical diagnostics and the development of virtual spectral databases for ultra‐low yield biologics and related molecules.

Hypothesis-and-Refinement Learning of Organic Structures from Multimodal Spectroscopic Data

Determining molecular structures from spectroscopic data remains fundamentally challenging because the inverse problem is intrinsically underdetermined: individual spectra are sparse, low-dimensional, and encode only partial structural evidence relative to the vast space of possible molecules. We address this challenge by formulating automated structure elucidation as a scalable hypothesis-refinement paradigm that tightly integrates spectral evidence with large-scale molecular priors. To supply structure-resolving NMR signals for multimodal learning, we construct \textbf{QM9SPIN}, a DFT-derived dataset comprising diverse 1D and 2D spectra, including J-coupling, DEPT experiments, and explicit spin--spin interactions. On this foundation, we introduce \textbf{SpectroMol}, a spectrum-to-structure model that proposes chemically valid molecular hypotheses conditioned on multimodal spectral inputs. Complementarily, we develop \textbf{MS-Mol2Mol}, a high-resolution mass-constrained molecular generator that integrates molecular formula, exact mass, and degree of unsaturation within a conditional generative prior trained on 400 million molecules, ensuring global compositional consistency and chemically realistic refinement. The integrated system achieves 93.8\% top-1 accuracy on the simulated benchmark, adapts effectively from simulated to experimental spectra with limited experimental fine-tuning, and further improves experimental predictions through mass-guided refinement, establishing a scalable route toward automated, data-driven organic structure elucidation.

High‐Fidelity Synthetic Raman Spectra Generation for Sinter Basicity Prediction Using β‐Variational Autoencoders

Raman spectroscopy has emerged as a powerful analytical tool across diverse industrial sectors, owing to its nondestructive nature, high chemical specificity, and ability to provide unique molecular “fingerprints.” In the steelmaking industry, this technique offers a promising route for the rapid and precise characterization of critical materials such as sinter, a porous agglomerate of iron ore fines essential for blast furnace charging. However, practical deployment is often hindered by the scarcity of labeled spectral data. This work investigates the use of a ‐variational autoencoder (‐VAE) for the generation of high‐fidelity synthetic Raman spectra from a limited set of real sinter samples (321 spectra). With and latent codes sampled from a full‐covariance Gaussian fitted to the aggregate posterior, the model produces genuinely novel spectra whose global distributional fidelity matches that of SMOTE (Fréchet spectral distance: 4.55 vs. 4.41), despite SMOTE being constructed by interpolation of existing spectra; the mean intensity correlation with real spectra is 0.930 0.038. These synthetic spectra were used to augment supervised regression models for basicity () prediction. The LightGBM model achieved (95% CI: [0.72, 0.84]) in cross‐validation with original data; no augmentation method produced a statistically significant improvement (Wilcoxon signed‐rank, all ), and on the held‐out test set (), no augmented model exceeded the original‐only baseline (; S‐VAE: 0.809; Interpolação: 0.786; ‐VAE: 0.780; SMOTE: 0.779). The operational value of generative augmentation emerges instead in a downstream classification experiment: classifiers trained with increasing fractions of synthetic spectra sustain higher macro F1 at 100% synthetic fraction (‐VAE  0.66, S‐VAE  0.60) than SMOTE‐augmented classifiers, which degrade to 0.58. These results establish that the value of VAE‐based generation lies in producing distributionally faithful yet novel spectra that remain useful when consumed directly by downstream tasks such as spectral library expansion and classifier training, rather than in regression gains.

LegoQ: Density-Matrix Representation Learning with Spectral-Spatial State Transitions for Hyperspectral Classification

Hyperspectral image classification is complicated by mixed pixels, spectral ambiguity, class imbalance, and limited annotations. Most current classifiers encode a pixel or patch as a deterministic vector and apply a linear or multilayer softmax head. Although effective for discrimination, this representation does not directly expose how mixed or uncertain a sample is. This paper presents \method, a classical density-matrix representation learning framework for hyperspectral images. The spectral bands are divided into groups and each group is mapped to a positive semi-definite, Hermitian, trace-normalized matrix state. A composable stack of spectral, spatial, and inter-group transitions then updates the states while repeatedly projecting them back to the valid state set. Instead of flattening the final features, \method\ aggregates the group states and compares them with learnable class-prototype density matrices through Uhlmann fidelity. The normalized eigenspectrum, von Neumann entropy, purity, and prototype fidelity provide sample-level diagnostics that are unavailable from a conventional vector head. On Indian Pines, ten runs yield an overall accuracy of $96.20\pm0.70\%$, an average accuracy of $95.57\pm1.29\%$, and a kappa coefficient of $95.66\pm0.80\%$. On WHU-Hi-LongKou, the best of ten runs reaches $97.52\%$ overall accuracy. Classification maps and feature projections show that the transition stack produces compact and better separated class structures. The results support constrained matrix-state learning as a practical alternative to vector-only hyperspectral classification without requiring quantum hardware.